Supervisory & Collaboration Team

Principal Supervisor

Clinical Collaborator

The student will work closely with Dr. Faradonbeh to ensure clinical relevance and applicability of the AI models developed throughout the project.

Introduction

The emergency department (ED) is a high-pressure, resource-limited environment that manages a wide range of acute presentations—from trauma and cardiac complaints to sepsis and neurological deficits.

Accurate and timely risk stratification is vital for prioritizing care, allocating resources, and improving outcomes. However, decision-making in the ED often relies on subjective clinical judgment, leading to variability and potential delays.

Artificial Intelligence (AI) provides a transformative opportunity by identifying hidden patterns in complex, high-volume data. This project develops and validates AI-powered decision support tools to improve risk stratification across diverse ED presentations.

Aim

To develop and evaluate AI-driven diagnostic support systems that stratify risk and predict clinical pathways for diverse patient presentations in the ED.

Objectives

1. Data Acquisition and Preparation

  • Use MIMIC-IV-ED, MC-MED, and HCUP NEDS datasets
  • Standardize and clean data across sources
  • Engineer features for clinical relevance

2. AI Model Development

  • Build models for risk stratification and outcome prediction (e.g., admission, ICU transfer)
  • Explore supervised learning, time-to-event models, and unsupervised clustering
  • Apply cross-validation, regularization, and sensitivity analysis

3. Clinical Validation and Contextualization

  • Evaluate discrimination (AUROC), calibration, and clinical utility
  • Benchmark against existing risk scores and physician judgment
  • Assess generalizability across diverse EDs

4. Dissemination and Impact

  • Publish in high-impact journals
  • Present at international conferences
  • Plan clinical translation and EHR integration

Potential Datasets

  • MIMIC-IV-ED: 425,000+ ED visits (Beth Israel Deaconess Medical Center, 2011–2019)
  • MC-MED: 118,385 ED visits with physiological waveforms (Stanford, 2020–2022)
  • HCUP NEDS: 30M+ annual ED records across 950+ hospitals

Research Methods

  • Retrospective cohort analysis
  • Machine learning model development & validation
  • Comparative effectiveness and clinical utility analysis

Planned Dataset Curation: RAPIDS

A customised benchmark dataset called RAPIDS (Risk Assessment and Prediction in Emergency Department Settings) will be curated.

Derived from large-scale sources (MIMIC-IV-ED, MC-MED, HCUP NEDS), RAPIDS will include:

  • Prediction targets: ICU transfer, in-ED mortality, prolonged length of stay
  • Grouped features reflecting real-world decision-making (labs, imaging orders, triage assessments)

This benchmark will enable evaluation of AI models on predictive accuracy, clinical utility, and time efficiency, supporting integration into ED workflows.

Timeline

  • Year 1: Literature review, data acquisition & cleaning
  • Year 2: Model development & validation
  • Year 3: Clinical testing, dissemination, thesis completion

Expected Outcomes

  • AI models for robust ED risk stratification
  • Benchmarking against clinical risk scores
  • Recommendations for integration into ED workflows

📩 Contact: Professor Massoud Zolgharni